1. Podcasts
  2. Operators
  3. Every AI Tool We’re Using & 4 Pillars to Guide You: Operators Build (E7)
Operators

5 Oct 2026 · 1h 1m · 29 chapters

Episode artwork

Every AI Tool We’re Using & 4 Pillars to Guide You: Operators Build (E7)

In short

The hosts outline a practical “AI-first” operating framework for a small DTC/e-commerce team, organized into four pillars: (1) a “company brain” (context + living memory), (2) an “operating system” (Winx OS) to centralize and visualize work, (3) buying purpose-built AI software/tools with APIs/MCPs (instead of building everything), and (4) using agents for repeatable workflows.

Guest backgrounds

No guest is clearly identified; the episode is primarily between two hosts (Operators Build, episode 7). They reference their team and partners (e.g., Motion, Skio, RichPanel, Runneth, OmniSend, Tribe).

Key claims

LLMs need company-specific context to avoid re-prompting and hallucinations; context must be continuously updated. Speed and workflow leverage matter more than headcount. Use a “10/80/10” approach (idea/AI execution/human QA). Prefer real UGC/human taste over AI-generated “AI-looking” content.

Notable examples

HQ as a shared “Google Docs” for skills/files + secure credential vault; Runneth agent auto-uploads ads, QA’s creator videos from Tribe, adds captions, applies naming conventions, and routes to the right landing pages. OmniSend MCP builds email/SMS flows. RichPanel automates customer support with a 50% automation guarantee in 30 days.

Written by AI. May contain mistakes.

Chapters

0:45

Four Pillars for AI Integration

Discussion of the four key components for integrating AI into a company.

“We'll, like, get into, like, how we're actually constructing the company.”
2:33

Understanding the Company Brain

Explaining the concept of a company brain and its importance in context engineering.

“So like one of the first things you did when we started on Winx was you set up Winx OS.”
3:29

The Role of the Operating System

Importance of having an operating system to centralize company knowledge and interactions.

“I think there's a lot of people out there that just think that it's stupid to pay for software in general now.”
4:00

Choosing the Right Software Tools

Debate on the necessity of using paid software versus building in-house solutions.

“We're looking for software that has an MCP or an API.”
5:11

Implementing AI Agents

Discussion on determining when to use agents and their benefits for e-commerce brands.

“Agents are a good place to start looking.”
6:28

AI-First Company Strategy

Exploring the advantages of building a company with an AI-first mindset.

“It's like 10 % of your effort goes into idea.”
9:10

AI Limitations and Human Touch

Examining areas where AI falls short and the need for human creativity and oversight.

“Like you and I will chat on Slack and then we'll have something implemented.”
13:17

AI as a Creative Amplifier

Discussion on how AI enhances creative roles while also addressing job value in companies.

14:01

The Importance of Role Value in AI Adoption

Discover how AI impacts the perceived value of different roles within a company.

14:55

Challenges of AI: Perception vs. Reality

Learn about the pitfalls of relying on AI-generated content and the need for critical evaluation.

Show all 29 chapters
15:41

Using Perplexity for Research

Explore how perplexity can serve as a research tool and the importance of validating its outputs.

16:41

Introduction to HQ and Its Functionality

Get insights into HQ and how it enhances collaborative workflows within teams.

“like pretty foundational to how it's also how we work together yeah is through hq we'll finally get to explain to people what it is.”
18:15

Enhancing Company Context with HQ

Understand how HQ helps maintain company context and memory for efficient teamwork.

“really like quality of life things that are great.”
19:22

Integrating Runneth for Workflow Efficiency

Learn how Runneth is utilized to streamline tasks and improve team communication.

“All I had to do is go into ads manager, QA it, hit live and it was done.”
21:42

Optimizing File Management with HQ

Explore how HQ organizes files and projects for improved productivity and context.

“when you say you have a local folder with HQ, you start working it, what does that look like?”
24:19

Sharing Knowledge and Context Across Teams

Discover the importance of sharing knowledge and context to build a cohesive team environment.

“Like I learned this the first time when I was building like a, maybe a CRO system for Jones Road.”
28:03

Giving Agents Context About Runneth

The hosts describe setting up their AI agents so Runneth and tools like Claude and Codex know about each other and the company HQ, saving context so they avoid repeated explanations.

“And the first thing we wanted was how do we get Runneth to have access to this?”
28:59

Building with Skiyo

Discover the benefits of collaborating with Skiyo for subscription growth.

“but we decided to work with Skiyo to build Winks.”
30:42

Exploring the Tech Stack

An overview of the tech tools utilized for operations and customer service.

“So like what's the rest of our tech stack?”
33:11

Runneth and its Applications

A deep dive into how Runneth is integrated into workflows for efficiency.

“Well, I like Higgs field because it's just the one I know.”
36:23

Optimizing Ad Workflows

Learn how AI can streamline ad generation and management processes.

“I believe, I don't know exactly everything it's trained on, but obviously Motion has just so much data and context.”
42:00

Design Workflow Challenges

Learn about the challenges in transferring design workflows into code-based environments.

“That extra 10%, what I wish I could do is hand it to a designer and have them actually do a step by hand, but there's not a great workflow for going because this is all to take it by hand now.”
44:42

Design Workflow Challenges

Learn about the challenges in transferring design workflows into code-based environments.

“T-Bar built the system behind all of them.”
45:01

Editing and Efficiency in Video Production

Explore how to improve video editing efficiency and team collaboration.

“So you can now go in there and say, like, we have consistent characters.”
48:29

The Intersection of AI and Creative Ads

Understand how AI can change the landscape of advertising and the importance of balancing quality and volume.

“It's then building like individual customer and individual SKU forecasting, which I think is a very interesting use case for AI.”
56:00

The Evolving Landscape of AI Models

Learn about the rapid changes in AI models and their implications.

56:55

Establishing Company Context

Understand the importance of shared memory in business.

57:51

Building Your Personal AI Brain

Discover how to create a personalized AI system for productivity.

“So that as you go through life, you're keeping a perfect copy of everything that you do with work and personal.”
59:19

Working in Goals and Loops with AI

Learn how to effectively set goals and utilize AI for tasks.

“And it's going to figure out how to reach the objective.”
A man walking through an airport, asking his phone a question

Ask about this episode

Get the gist, explore an idea or ask about something said.

Claude

On Operators, what are the key points of this episode?

Searching your listening

Transcript

Automatic transcript. May contain errors.

0:00Matt Bertulli: Okay, we're here. It is episode seven of Operators Build. This is the all things AI episode. I want to start by saying I think there's like four big things that we've created that I think are like pre-universal everybody should do. So the first is you have to think of company brain, right? This is like context, memory, all the stuff that like the models are just bad at. probably the most important part number two would be like an operating system so it's nice to use all these tools it's great to build software but you need some way for the humans in the company to interact and visualize things right so like we call that winx os agreed number two all right number three uh all of the software and tools so i don't know about you man but i i'm just a believer there's a lot of stuff that we shouldn't be building that we should just buy and we should just use so this is like the omnisense the rich panels the iris high beam like all the the tools that we've opted to buy for the company some of these are sponsors some are not and then number four uh we'll call them agents so for winks we have like a researcher a brief writer and we probably have other ones like competitor analyst yeah those are kind of the four big buckets and i think what we do today cody is like go very deep on each of them um and we even show people Like, let's share a screen.

1:21We'll, like, get into, like, how we're actually constructing the company. All right, let's get into it. Explain brain to people. The brain is your context layer. And so context engineering is one of the most important things. With these agents, with the claw to codex, they're the smartest, most capable things or people ever. But they're not trained on your specific business. They have a lot of general knowledge. And what you need to do is equip them with specific knowledge. And I think about it the same way that you would want to educate somebody if they were coming into your business. What's all the context you would want them to know and have access to?

1:55What is your mission? What is your vision? What are your past tests? What are your products? What are your cogs? What are your SKUs? What are your reviews? What are your customer insights? You just want that dumped into one place. And then these agents and any tool, AI tools that you're using will then be able to access that and have knowledge of it. So you won't have to prompt it every time and say, hey, this is my company. This is what I want you to do. It'll just know what to do if you have the right setup.

2:22Matt Bertulli: And the thing with this is important is it's updating all the time. Absolutely. So you don't like set something like this up once. It's a living, breathing, like that's why you call it a company brain. It's like you have to have it. Always. Yeah. Let's go to number two then, like the operating system. So like one of the first things you did when we started on Winx was you set up Winx OS. What was the thinking? Like, why is that important? I would say this is one that's probably like not the most important. Things just live in so many different places, right? And like we would be getting ready for a launch and we would need, you know, cogs from this place and we would need shade descriptions from this place and copy here.

2:55And it just takes so long. I think the big theme is just everything in one place, whether that's for a human or for an agent, have everything in one place that it can access. See, I disagree.

3:03Matt Bertulli: I think that it is actually really important because if you've ever used one of these LLMs, like to people watching, listening, they aren't very fast. So if I had to say, like, can you show me the last 10 ads that we've made today? That's going to take a long time. Whereas if I go into Winx OS, I can just click and see them. They're there. So there are certain things that like you should not be using all of this compute power from these models. So I still think it's like more important. Okay, let's move on to like software. I think there's a lot of people out there that just think that it's stupid to pay for software in general now.

3:34Where do you stand on this subject? Because like we're using a bunch of software. Same thing. Just because you can doesn't mean you should. You know, it's really easy to feel productive and it's really easy to waste time doing it. And there's a lot of maintenance costs and stuff like that. Yeah. Like I think when it first came out, it was cool to try to build this stuff. Oh, yeah. But now it's like, no, like spend your time on the most valuable parts of what's actually going to grow the business. And we're a two to four person team. We don't have time to waste.

3:58Matt Bertulli: No. You know? Yeah. Yeah. So I think for software, we're looking for a few things. We're looking for software that has an MCP or an API. It's almost a non-negotiable where we can use it because there's a lot of value in taking the context and the brain and then giving it the tools to go and use a specific MCP to get access to something. And it can get more information about the specific tool, either that or a tool like a rich panel or an iris where maybe it doesn't have the MCP, but it's purpose built for that one specific thing. And like experts at technology and building softwares and finance or CX have actually built it to do that function.

4:36And they're going to do a way better job than you and I are going to be able to do. Yeah, I 100 % agree. Okay, then the agents. When do you decide that something should be an agent? I truly think everybody listening that has an e-com brand should be using agents. I mean, if you're using Cloud Code, that's a coding agent. You know, Codex is a coding agent. It can go do stuff. You can watch it use a browser. So that's very basic level, but you know, something like a runneth or a Hermes or something like that, or a Grokbot can be really, really powerful to just allow you to do a lot more with the leaner teams.

5:07Like scheduled tasks, workflows, like things that are just going to happen a lot.

5:11Matt Bertulli: Agents are a good place to start looking. Definitely. Yeah. When we set out to build this company, one of the things we said from the beginning was AI first. Maybe tell me, what does that mean to you? when you hear like, oh, I'm going to build this company AI first? We're in a good time for us because we don't have people. And it's very hard to go AI first at an established company where you have like, instead of tech debt, you have like people debt and you have like workflow debt. And it's really hard to get things to change. And so the beginning, it's like, we don't have people. Before we hire somebody, before we do emails or do ads, like let's see how far we can take it.

5:46And it doesn't have to be this binary like people or not, but it's like, how do we give ourselves more leverage? because it's essentially you and I day-to-day in marketing and we've got ads to run, emails, landers, all that stuff. So it's like, how do we just give ourselves leverage

5:59Matt Bertulli: to do as much as possible and do it to a high level? Yeah, and what I'm noticing in other companies, like coming from Jones Road, coming from Pila, it's actually the workflows that are a hard part. It's like you have all these SOPs, so many systems, and they were all designed like human first. And the way that we're building things is whenever we have to do something, we just start with, okay, well, can we get AI to do this? Can we start the workflow with AI so that we're only really reviewing things? So like I would, I think this framework of like 10, 80, 10 with AI is great. It's like 10 % of your effort goes into idea.

6:35Matt Bertulli: So like right now we're talking offer yesterday. So it's like idea. Okay. Now we're going to get AI to do 80 % of it. All the grunt work. That's the workflow. And then the last 10 is let's make sure that it didn't hallucinate and do something really stupid. Yeah, yeah. And it always does. So like there's always like just dumb in the output. It's definitely not perfect. Oh, God, no. So why do you think this is important to build a company this way? You know, just the way DTC is going, e-commerce is going, like more of your P &L is going to go to CAC as possible. And so the best companies are just going to be extremely high leveraged and they're going to have, you know, OPEX 1%, 2%, 3%.

7:15Yeah. It's just like D to C 4.0 model. And it's just probably not possible to do without a ton of automation and tech. But there are, there are 50 million, a hundred million dollar brands with like two, three, four full-time people. It is. I have a lot of questions on whether or not those companies are sellable.

7:33Matt Bertulli: Current market's sellable, right? I think that there's like the, and this is a conversation that at some point we'll probably continue to have it as a bunch of shareholders. because I think there's an argument right now that the market doesn't know how to buy a business without employees. It's sort of, I don't know if you remember, maybe like six, seven years ago when the whole like VA thing or the whole like offshore thing started, same thing. There were buyers in the market saying, if there's five of you in America and a hundred of you somewhere else in the world, how do we value that company?

8:04Matt Bertulli: I think we're there right now. My bet is with what you're saying is that fast forward five years, we will absolutely have a framework as an industry for how to buy a company with one or 2 % because it'll just be normal. We'll know how to underwrite these deals. Like private equity, we'll figure it out. So this is sort of like a, for me at least, it's very forward looking bet on like, this is just what the company of the future looks like. Yeah, I agree. Especially a consumer e-commerce brand. Yeah. And then the other thing is speed. Like when at Jones Road, when I was doing stuff and I did a lot of web and CRM stuff, you know, you can do a lot in these roles or conversion or design and development, strategy and copy.

8:44One person can do it all. And so, yes, you need less people, but also as soon as I finish strategy, I can go to design. As soon as I finish design, I can go to dev versus normally you have to do a handoff, right? Just more people involved slows things down. And I just think fundamentally speed is the biggest competitive advantage in a business like ours. Like we need to go from how do I have an idea? How do I have an insight? And like, how do I get that live today? And like, that's what the best brands are doing. whether it's AI or not, that's what the best brands are doing.

9:10Matt Bertulli: Yeah, we've already seen it. Like you and I will chat on Slack and then we'll have something implemented. It will be ours. Whereas if five years ago, you can have a conversation and it might be a week, two weeks, like stand up a landing page, build up something new. Like we do this in minutes. We launched with seven landing pages. Maybe they weren't perfect and great, but I remember when Jones Road launched five years ago, like you wanted a landing page, it was like 20K for an agency. Right, yeah, it's a very different. I also, I think what you're talking about, so I didn't expect this with AI.

9:42Matt Bertulli: My initial expectation with all tech is like, this is going to flatten an org. So like companies themselves, like we'll just flatten them out. We'll have less hierarchy. What I didn't expect was that the roles would also start to merge, you know, like designers and developers. That's just like a person now. So I think that's also interesting is like neither one of us is like super technical by trade two, three years ago. but if you look today it's like we're doing a lot of code commits we're doing a lot of everything like your github usage is way up we're designing i never designed before i can't design save my life i basically only wrote copy yeah that was it uh now it's like you can take copy and turn it into any asset and i think that's kind of what we're seeing so what we're going to get into is like what functions and departments are like what are we actually using ai for but maybe let's start with what are we not using it for because that's going to be a small list but it's probably an important list.

10:36Matt Bertulli: So like, what are areas that you think we just don't use AI to do these things? There's probably a few, there's probably a lot that are like not black or white. Like we're using AI for a lot of copy, but in some ways AI sucks at copy. So it's not like we're just trying to one shot it there. No. Design, we're using it a lot. Our first employee was a designer slash editor that we brought on. Like there's a lot of limitations of that. Yeah. We are using it for images, assets, but I think there are pros and cons of that. There are some people who are incredible at making AI really not look like AI images.

11:13And those are actually creative people. I actually don't have the skill set, not in the AI skill set, the mood boarding, the creative skill set, but I've seen some people do it. But I think we want real images. We want UGC. We want real people. yeah yeah real humans so i think people and i don't know that we have a hard rule where it's like no people but it's there so i don't think we want that but like we want content like i'm we'll do some ai ads but i'm not bullish on like ai ugc like we want real content so how do we you know for our creator programs how do we do all of the how do we automate all of the back-end stuff the uploading of the ads sending briefs and stuff like that the analysis like let's automate that so we can actually get like real content.

11:55Matt Bertulli: Yeah, I think there's like a human taste element that AI is terrible at. Yeah. And that you still need people to start every idea, right? Actually, I don't know. I use it to help me ideate. I definitely do. I'll start, if I have an idea on copy, I might go to Claude or ChatGPT or whatever and get it to give me like 50 or 100 variations of the idea. but then I quickly narrow it down and I try to write most of the copy myself. Most AI tools right now are all promise and no delivery. You know exactly what I am talking about. Super fancy launch videos. This is why I'm loving what RichPanel is doing.

12:38Matt Bertulli: They have AI and support smashed together, made it practical, made it useful, made it valuable to brands today, not on some future promise. And instead of asking you to spend weeks writing prompts and uploading help docs and babysitting, they flipped the whole script. Their AI builds your support team for you and everything it needs. I've watched it. It works. It's freaking amazing. RichPanel even guarantees 50 % of your support volume will be automated by AI within 30 days or your money back. I call that a no-brainer offer. If you are interested, go to richpanel.com slash demo, not for some generic demo, but to actually book a call and watch them build your actual support team.

13:16Matt Bertulli: i don't know like how nerdy on copy you are but we used to copywriters used to do this thing where to get good at copywriting you used to actually just hand write yeah handwrite sales letters and stuff you just like take it and like i would just sit there and i would copy out copy copy uh and that's how you learn to do copy i think that all ai is going to do is like make great creatives better creatives make great copywriters better copywriters um it's going to be like the great amplifier i completely agree with that i had that thought when you were kind of describing it yeah completely think about think about the same way and like what i've seen is the people that are very good at their subject matter matter expert they are much better at it or like they can do more like the creative can actually design stuff in ai way better than i can yeah but i think here's here's the difference not every role in a company so this applies to consumer applies to everything your business doesn't need superstars in every function sure so i think in consumer we need really great creative uh that's like image video copy and we need really good sort of judgment everything else is secondary and this is mean again if you have employees no employee likes to hear this it's like no your job's just not as important but it's true like some some roles are just less valuable than others so i think what ai is doing is you get people who are amazing copywriters well now they can be a mediocre everything else and that's fine because the company doesn't need them to be amazing at those things you can just be a mediocre supply chain person because the the super brain is going to be way better whereas copy well no you still need to be a pretty freaking amazing copywriter you'll be very good at making ads that's kind of what consumer is and there's some again areas we're gonna like you said it's a 1080 10 we're gonna you know we're never just kind of one shot it like yeah even like research and stuff like that like you just have to be careful and have to know what you're looking for yep because it'll you'll get slop and you can get like I've heard people talk about this where AI can give you like the perception of competence yes but it can just give these long reports and stuff like that like you have to know what you want it to do and know what to look for and kind of curate it down I used perplexity a lot to go do research for Wings like persona research and it would come back with these like crazy ass briefs and then i would ask it for sources and i would go to like reddit or wherever and like read the sources i'm like that is not at all what i i read and i'm like what you your your inference you know as an ai is completely off the mark of what like these ladies or this guy was saying so i think that's a good call everybody should be very careful of just like blanket following this thing um it is really really really bad bad at constructing um video scripts i find and maybe this is just how i use it i find it terrible at like building a narrative you know it's sometimes it'll like it'll get one or two lines right but i find the actual structure just never looks quite good and i got skills for this i've built like built so many versions it still is just not there whereas like if you meet somebody who can write a great piece of content that it's like it's so natural to that person so i just find it's not great there okay let's do this why don't we show people uh like our entire stack because i think that'll actually be helpful how do we work like let's start with hq because i think that's like pretty foundational to how it's also how we work together yeah is through hq we'll finally get to explain to people what it is.

16:52The way I describe it, and I don't know if you have a different way, is like normally AI is like Microsoft Word or Excel. It's like if I make a change to a financial model or Word document, I have it. And if I want you to have it, I have to send it to you. And then we can't really be working on the same thing at the same time. And that's like super outdated, right? Obviously everyone is using, you know, either you can use a SharePoint or if you're using Excel or you can use, you know, Google Drive. But that's how most AI is. Most people are doing things locally, right? You have files on your computer, you know, and then you have to go and share it.

17:25And there are definitely ways without HQ to do it. We did a Jones Grove before HQ, but you have to use GitHub and push and pull and teach everybody that. And so HQ just makes it really easy to be able to make AI stuff multiplayer. So it does a bunch of things, but essentially any file that I have and, you know, a skill is just a file, right? It's all Markdown files and stuff. When I'm working in my HQ, which I only do, my local HQ folder, or I just have a folder, it kind of automatically gets pushed to the cloud, right? And so it's just, you know, shared and then automatically it's pulled down to yours, to anybody who's in our organization has it.

17:59And so we can be working on the same thing. So it's, you know, my analogy is it just takes it from Microsoft Word to Google Docs. And it's not quite real time where we can't do the exact same time, but it does help. So any skills that we have, we can just share them automatically. You know, and then there's just a few small, really like quality of life things that are great. So like the skills, any like integration, MCP, you know, because I'll share credentials. Exactly. Because I'll do, you and I both use Codex. We use Claude. We both have a Mac mini. We have a Hermes, stuff like that. And so I used to have to put them in three different places.

18:33And then I, and so there's just one, it's very safe, secure. It just opens up a tab, you know, a window in your browser and you put it in there and it's very safe. Because I've prior to that wasted a ton of time trying to, you know. Oh, yeah.

18:47Matt Bertulli: Yeah, like how you access Shopify is how I access Shopify. Yeah, yeah. OmniSand, OmniSand. Like we have shared credentials, Higgs field, whatever we're using, HQ has like the secure vault of like all those credentials. Yeah, the way I explain it to people is, yes, it solves the multiplayer thing. So how do you get that? For me, I mean, the greatest thing with AI, the greatest downfall with AI is like, we all have a memory problem, right? And we have super geniuses with amnesia. So how do you solve the amnesia thing? there's lots of ways to do that locally people use obsidian they create like you know whatever your own g-brain like all that stuff but company memory that's the issue so i think hq solves that um and then credentials like those are sort of the three big things you can do other stuff in it but like really for how we were working it's mainly those three context yes context yes company context which is one of the most important things like dude it's like all company value is inherent knowledge well and and like you know two years ago whatever it was like people talked a lot about prompt engineering like yeah that's that's pretty much that and all i we can share how we're actually using the models and stuff but like all the models are so good like they don't really matter much no prompting is that it's really context and then like understanding how to set loops and goals and stuff like that but you have to have the company context you know i think that's it where people will try to write copy others try to do something that's like not good but you haven't built the skills you haven't built the context and it takes a ton of time i don't know how much time you put it but like it's a lot of time and they're never ending like if we build a skill it's like we're constantly refining yeah but then once you do it and you you marry that with the context and the integrations of the different mcps you're doing it it can get really powerful i wasted two hours over the weekend trying to upload ads and i said i never got to do that again so i had run a do it looked at every ad picked the right landing page it named all the ads, wrote the headline and copied to match from options I gave it, set the name and conventions, set UTMs, all of it and push it to ads manager for me.

20:47All I had to do is go into ads manager, QA it, hit live and it was done. Runneth is an agent with its own virtual computer. So it never shuts off. It's always on. The whole team can talk to it. The same Runneth and Slack. It remembers everything. It has prebuilt skills based on some amazing stuff that the motion team has built. Here's one example of how I'm using it. Last night I hooked up Tribe, our creator sourcing tool. And anytime a creator video comes in, it gets QA'd against my feedback that I gave in it before. We get notified of any new video. I can have it add captions. It watches the video, figures out what it's about, adds naming conventions, edit hooks.

21:23I've been doing a lot, building a lot of static ads with Runneth as well. It's pretty awesome and it all comes out of the box. We've been using our Winx as our main creative strategist, using it to build tons of ads for us. Winx is using it. Ridge is using Runneth. HexCloud is using it as well. Go to runneth.com. They're giving listeners 500 million free tokens for your first month.

21:41Matt Bertulli: So can you show people, when you say you have a local folder with HQ, you start working it, what does that look like? So when I am using Codex, which I use more often, but it's kind of just my default folder, right? And so normally it's just working on desktop or working on a folder and you just give it permissions. But now default, I'm really only ever doing things in HQ. And so this will just have context of really all of the files that are in there. Okay, so context. You have a HQ folder. You start all of your work, you start it in this one folder. Yeah. So instead of having, let's say, five different GitHub projects, one for Shopify, one for OmniSan, one for...

22:22This is just one. Inside of there, there's different companies. So there's like, I have one for Winx. Yep. I have one for like my personal. I have one for like, you know, my, like for podcasts, stuff like that. But yeah, you know, so for Winx, there's one folder and one company. Yeah. And so any work happens and goes in there.

22:41Matt Bertulli: And then in that one company, Winks, we might have like, we have a GitHub repo for Shopify. We have a GitHub repo for something else. So then it's all inside of that one company. I would say like GitHub is like slightly different where like this is like the context and stuff. It's not, doesn't house the actual code base. Right. So the actual code base is in there. Yeah. But it kind of has knowledge of everything. Yeah. It can see it all. Yeah. Right. Yeah. So if you're going to go work on, like right now, you're probably doing work on our Shopify site. You'll still start that in HQ. Yeah, yeah.

23:09Even though there is still a separate GitHub repo for Shopify. All of our skills live in HQ. Yep. And so even prior to HQ, if I was working on our website, I don't want to do it directly on our live site, right? Normally you have a dev theme, but I actually would be working on that locally on my computer. No one will be able to see it. And then I would push that using the Shopify CLI, the command line interface to Shopify to be able to preview it there. And then once I wanted to actually, we'll call it submit that or commit it, I would open up PR, a pull request and have that go to Shopify. But so it's just I'm doing it locally, which is HQ, but then it just syncs it with like the HQ cloud.

23:49Matt Bertulli: Yeah. And so the benefit of that is A, it gets saved. B, it's again, it has all of that context in one place. So as I'm working, I'm working on an ingredient section right now and I'm like, hey, pull from this claim, pull from the study. because what I did is, you know, I just think the context of engineering is really helpful. You know, we have Dr. Todd Bourne who formulated a bunch of stuff. He tossed in a Google Drive folder of all studies. So as soon as I see something like that come in, just give it to an agent and say, hey, toss this in HQ, this is what it is. So it just has the memory of that.

24:20It has a context of it. And so it's just really powerful. Like I learned this the first time when I was building like a, maybe a CRO system for Jones Road. I built a copy skill and copy context for a different project. and this was when I was earlier on and the CRO skill was just like calling upon it and writing copy as I was building landing pages even without me trying to. Yeah. So like sometimes even when you don't know the intended goal, just getting all of this context in one place can just be extremely helpful and valuable. Yeah. I think it's the thing too.

24:50Matt Bertulli: So obviously like for the show, we have different sponsors for different things, but we also have a lot of other tools that we're using. So like Codex isn't sponsoring the show. Claw's not sponsored. Like this is just our chosen stack right now. Like we use Higgs field for things. We use Milius for things. Like we're trying out to, like I think my approach at least has been like, let's just try out everything. This is a fun time. It's nice when you're starting something scratch. You're like, screw it, man. Let's just try it all. Like we don't need to be married to anything. In this case, like HQ, I'm actually super impressed.

25:22Matt Bertulli: Like I was a bit hesitant when we started using it. I'm like, I don't know. This seems like onerous and more. I had something else I had to learn. I didn't want to do it. yeah i'm kind of happy now um it was like a pleasant surprise that's with all these things it just takes time to learn yeah not all of them are worth it but yeah but i definitely think you know it it allows us to to share stuff and not have to do this like i think the start work was it the handoff handoff so like if i'm working on something and i want you to have it i can just hand off and then send you a link yeah just tell me what the file is yeah and then it just keeps working i i think for the people who are listening or watching and if you're going to be doing work together as a team this is whether you use hq or not like just figure this part out yeah as productive as you think you're going to get single player the multiplayer thing is where like the company part gets built definitely and and the brain and like the company oh yeah the company brain thing is like is huge let me let me show it if that's helpful let's do it cool so i'll share my screen here like this is what it would actually look like so i'm inside hq here again i i never have to go in here but just to see what the you know organization you know system is so this is cool like people have probably seen this with uh obsidian yep but it's a very similar thing where it just shows you all of your your different things you have and here are all the skills that you know we have and kind of the the connection they have with each other these are a lot of the different files that you know we've built and kind of projects that we've done um knowledge these are just various this is just our brain but these are just you know yeah different files competitors you know different things that we've done over time best practices like the same i i think about training agents and building context the same way you would think about training employees and onboarding employees just here's everything i know here's how i want this to be done um and so that has all the be it you can see if we hover over it it this is a good frame because like you every employee you hire needs context like you have to get them you onboard them yeah you're on board them yeah teach you about company culture like all that stuff so like this is sort of where we do that and then people can see like there's a folder structure that is in place right there's projects there's skills there's just like lots of stuff and everything goes in here i want people to get like we don't work outside of hq so even when we use like we're going to do work with omni sent right we're going to work with the omni omni send mcp to like build emails something We do it through HQ.

27:44Yeah, the emails live in HQ. And then we use the OmniSend MCP to push it to our OmniSend account. Right. So here's knowledge. And this is the company brain, but just the file structure of it. And you can kind of see what are new things. And so one of the things, one of our sponsors is Motion and Runneth. And we're using that for agents. And the first thing we wanted was how do we get Runneth to have access to this? But also we need Runneth to know what HQ is. So if we tell Runneth, hey, go find this. but vice versa we need we need our agents to know like our clawed and codex to also know who runneth is as well so this is just context just you know you just go in and say hey this is this is what this is this is what runneth is i'll even go to runneth and say hey this is what i want to do please write up a document that i can copy over to explain what you are and who you are yeah and it'll just save it and so you don't have to tell it again you know that that amnesia thing

28:44Matt Bertulli: You have been listening to us talk for years about building nine-figure durable goods brands. Hexclad, Ridge, Simple Modern, Pila. When we sat down to build Winks, there was never a question it had to be a subscription brand. And being the veterans we are, we could work with anyone, but we decided to work with Skiyo to build Winks. Working with the team over at Skiyo has been awesome. We get to see all of the stuff that the fastest growing CPG brands are actually doing to acquire more subscribers and keep them longer. My favorite example of this has been incentivizing new customers to go from a monthly subscription to a quarterly subscription right after their first purchase, but before they get their first order.

29:23Matt Bertulli: The team at Skio has been sharing insights on how much in-portal flexibility to give subscribers, when to change frequency, when to offer product swaps, when to do gifting, what rewards should look like, what is cancel flow optimization. The list is freaking endless and they have shared it all. They have been incredible to work with they are bringing so much value to this partnership almost feels a little unfair if you run a subscription brand you should be talking with the team over at skio they know their stuff there's just you know there's knowledge and um the the folder structure which i don't you know necessarily look and care about but just of having the different kind of folder structure the same way you would do like a google drive kind of where you have yeah you have a department and then you have subfolders within it and what's you're right here like i actually don't look at what's in here right i use codex or quad to look at what's in here um i also have i don't know if you've done this but like i have uh i will take like my hermes agent which is my personal like a ceo brain thing and i will then have it commit things and it can work with hq so it's just like anything that's like multiplayer it's like okay just put it in here because like Like we need every person that we add to the team needs to have access to this.

30:34Yeah, just dump everything in. Okay. That's my philosophy, yeah.

30:38Matt Bertulli: Okay. So then from HQ, let's talk about like what have we plugged into this thing? So like what's the rest of our tech stack? Let's just go through that for people. So we've got OmniSend for email and SMS. And so what's really cool is we're doing that from Codex, Claude, Terminal, something like that. Yeah, like we don't log. Like I don't, I log into OmniSend simply. It was so easy to just like MCP and build the emails and then use. Yeah. That I actually, the only reason I logged into OmniSend, because I didn't believe that it just. Oh, just to approve. Yeah. Yeah. I just like, I need to see that this is true.

31:12Matt Bertulli: Yeah. Then I would go in there when we were building the initial flows. You're like, can you just go in and check everything? And I'm like, I would go to HQ and use Claude. I'm like working. I'm like, no, I need to go double check that this is all actually accurate. Oh no, there it was. It's pretty cool. Yeah. That's pretty cool. Yeah. That was my other big surprise was like how easy and how good the OmniSend MCP was. Like, well done to that team. Okay. So there's OmniSend, RichPanel, right? So like RichPanel for customer service, AI, like they've been a long time sponsor. Yeah. RichPanel is a little different because we're not doing that from like a clutter codex.

31:46Matt Bertulli: Yeah. That's just part of our AI stack. Exactly. We just need best in class, like customer service. Yeah. But it's like solved, handled, like very rarely do we have to go in and - think about it. He's got to go check. Okay, so then all website work, right? Yeah, we don't have a developer. We are the design and dev team currently. All that is just done through HQ. Are you using Codex or Claude more for that now? Primarily Codex. Is that an Astra thing or is that... Even before that, Codex was terrible at design prior to GPT 5.6. So it was definitely more Claude before that. Fable's really good at design, but on 5.0 it's just eight tokens.

32:23It's way better now on 5.1. And so I do use it more, but Astra is pretty good at design. So that's my preference. Do we have like a set of design skills for the brand?

32:35Matt Bertulli: And is that an HQ? Yes, they all have an HQ. So we have design system. That's like one of the, you know, really important ones. And actually I can share like Winx OS as well. Oh yeah, we should show people. But we have design system. We have skills for development. Like we have very specific rules for like never push to production. Here's the audit, you know, here's different things like that. So we have, there's like a whole workflow that again has been refined that allows non-technical people like you and I to, to, you know, do our website with no developers. Without also breaking it every time we touch it.

33:07Matt Bertulli: Yeah. Okay. So websites done through there. What about asset generation? So images, videos. Yeah. So you like Higgs field. So you use that a bunch. Well, I like Higgs field because it's just the one I know. Yeah. Old dog. And yeah. But I think we've got Mellius. And then we use just like open AI API as well. Yeah. I built a skill. I think you had some of your own as well of, you know, doing that. So we kind of built some skills for like static ad design and things like that. Planner obviously with like the videos, the animated videos. We've built some skills. Yeah, we had Sean showed us how to make animations.

33:40Then we went to town.

33:41Matt Bertulli: Okay. And then Runneth. Let's talk Runneth a little bit because I've actually had a few people message me. How are you guys using Runneth? Where does it live? What is it good at? Maybe you've been, I watch you in Slack and you just boss it around. I think I'm nice to it. Yeah, you're very polite. But what are you mainly using Runneth for? It took me a while with Runneth and even OpenClaw to understand all of it. So I'll try to explain a little bit. You have really, I guess, two different components of an agent or anything. And you have the harness and the model. Yep. Right? And so the model is from Anthropic, you have Claude, you have Fable, you have Opus.

34:20There's the different ones. That's the model and, you know, there's different sizes and token cost of them. And then you have the harness and the harness is essentially like the UX that you're using it with, with like any like prebuilt skills or things like that. Oh, and then you also have the machine. Right. So like we're doing stuff locally on our computers primarily. But if your laptop is closed, it's not running. So it's not, you know, always on. And if you have automations when it's closed, it'll stop. So a lot of people, what they did, you know, with OpenClaw, which is a harness that can then run on a model.

34:54And yeah, I think we have our Hermes as well, which runs on, you run yours on your codex plan, right? But Hermes, OpenClaw are two other ones that require a little bit more setup. They're like very early and they kind of, you have to be a little bit more technical. And most people run them on a Mac mini because it just can always run. It never closes. And so that's what we do. but it does require like more setup and there are definitely like pre-built skills that come with those sure but they're very general right like there's a skill with harmonies where it auto learns and creates stuff but like there's no marketing knowledge that it has there's no yeah you know company knowledge what runneth is is again it can run on multiple models so it can run on you know all of them we have access to them because they have different skills yeah um it has a lot of pre-built things for example you and i spent a lot of time building static ad skills with Runneth, it was just done.

35:42You know what I mean? It's actually good. It's really good. Yeah. Because they've trained it, you know. It also runs on a virtual computer. So instead of a Mac Mini. Yep. And now, you know, I think they were pretty early to that. Rezo was early to it. Now GrokBot is, you know. Yeah, I know.

35:55Matt Bertulli: They're all kind of copying it. They're on a virtual one. Muse is on a virtual one. Instinct is doing the same thing. Like they're all giving you like sort of dedicated hardware because they also realize like you got to solve the memory problem. Exactly. So that's essentially it. And then again, you can use it where you want. You can use it in, you know, the Motion app and run it or in Slack. And so I also think it's really cool just to have it in Slack and collaborate with people on it. Yeah, workflows. Like we live in Slack. Yeah. Like most work in companies is like in a Slack. Yeah. Yeah. Yeah.

36:25I believe, I don't know exactly everything it's trained on, but obviously Motion has just so much data and context. You know, like Alicia, who's great, who's working with us. You know, she's got a lot of different just like creative strategy stuff. Yep. And so we have it in HQ, but it can do a lot like off the shelf. I'm getting back into the weeds. One of the things I'm probably the rustiest on is email. I'm super excited to be using OmniSend. They're using them on Winx. We're AI-pilled. We are using it to do everything. I'm running most of our email program from Claude and Codex using the MCP, and it's awesome.

36:57We also have email, SMS, and pop-ups all in one place. We have all of our flows, welcome series. It's all talking to each other. I love the platform. The analytics are really good. to highly recommend it, especially if you're in the position that Wings is in, new brand starting out. But if you're on a different platform and you're paying too much and you want to switch, they will do the migration completely for you. No lost data, no lost list. It takes only five days start to finish. And they are giving our listeners 30 % off the first three months. Go to omnisend.com slash nine operators to try it out for yourself.

37:30I have been very impressed with what it can do. I spent two hours last week uploading ads. and I'm like, I'm never doing this again. And so Runneth is uploading our ads now. So any ad, it'll watch the ad. We create it again. You just have to build this stuff from the beginning.

37:46Matt Bertulli: You can see like it's connected to tribe. So like our tribe creators submit videos, right? And then you can have Runneth upload the videos into. It does that. I had to add captions. Next step is working on video editing skills. It'll watch the videos, for example, for a static ad or for not. It'll figure out what they're about. So we built a database in HQ of our landing pages, right? And if it's a more general sleep one, it'll name the naming conventions and it'll pull copy for the headline and text based on what we've given it and it'll run it to a general landing page versus if it's a melatonin one and it figures a melatonin, it's gonna go to our melatonin page we told it to go to with a melatonin headline.

38:29That's another example of like the workflow debt. But like at Jones Road, we get a spreadsheet and there's somebody there who's sitting there matching these things up and going like, and it just takes a ton of time. And so now it's automatic. I was at the US Open the other day. We got some tribe videos in. They got sent in Slack. I approved them, Runnith built them, you know? All you have to do and go into meta and just, I asked Runnith to pause them, keep uploading pause, and then I go in and upload them. It's the 10-8-10 thing, right?

38:56Matt Bertulli: It's like, look, we go in, we recruit the creators. We do all that stuff. Mm-hmm. Then there's this workflow that happens, right? Creators submit videos, run with goes in, gets them, sends them to us. We approve, uploads, we approve. That's what I mean by like, that's a very AI first workflow. That would have, and my last company, same thing. It's like, there would have been a creative key sheet where like all of this stuff is in there. It's all tracked. We try to naming convention, match it to meta. Like we don't do any of that. None, none. And it's, I was thinking about this before. it's all about how ambitious you want to go.

39:32We're pretty advanced. And I was surprised that Runneth could do all of this. And so now even the motion team is pushing us to be like, let's integrate more. What feedback does Cody normally give on tribe videos? Let's have Runneth learn from that. Can Runneth start giving feedback on the videos as it comes in?

39:47Matt Bertulli: So what are some examples? You've got Slack open here. What are some of the things that you've done with? So there's one of our tribe videos. So we just set this up where just any new tribe video comes in. Again, this is a pretty basic one. but any new tribe video comes in and just you know to be able to watch yeah um i'll show you i want to create you know some some ads i mean this is one i do all the time we're like i don't think that this is the most sophisticated way to come up with ads but i do it for fun because i'll just be scrolling on x or on social at night and i'm just like hey make this for us and again yeah play with the site like this idea i like this idea can you make me a version and that's all and and because this shows, right, if we didn't have the context with HQ and all this stuff, it's like, hey, I want you to make this ad.

40:31My brand is Winx. We target moms. It's like, no, like, here's our ingredients. Like, you do it once, you put it in HQ, you have context, and then you say, make it for us. And it's not always going to do it perfect, but it knows, because I've asked it to do it before, always make me a four by five first. Once I approve that, I'll do that. It'll analyze this one, right? It'll analyze this, you know, ad. It'll pull that. It's trained on a decent amount. And it made me not. I mean, is it perfect? No, I don't think so. But I'll show you. I mean, it did pull, again, without asking it, images of the ingredients.

41:05It had context of the ingredients. But like, there's some small design things. Like, there should be more spacing there. There should be more padding. Yeah, sure. We can improve the copy.

41:13Matt Bertulli: Like, I think it's just like you go from an idea to a prototype very quickly, which a lot of people are using AI to do that with software. I mean, one of my goals is like, I want to build less software. Like I don't actually want to use AI to build a lot of custom software. Custom software needs to be taken care of. Yeah. It's like saying I want 45 kids. Like I don't want 45 kids. That sounds like a lot of work. I would like to just continue to use the models, right? And use something like run it. So I'll have it get better. So now that we've got this, to me, it's like, great. You can take a designer and just like, this is actually really good.

41:43Matt Bertulli: Like clean it up. Yeah. And make a better version now. That is the one part where the handoffs are not good. that I wish. So on that 108010, like I can one shot, maybe I'll do this. Like I can one shot a landing page and get it. We'll call it this like 80 % of the way there really easily. That extra 10%, what I wish I could do is hand it to a designer and have them actually do a step by hand, but there's not a great workflow for going because this is all to take it by hand now. This is all code. And so it doesn't transfer super well to Figma. You know, cloud design is decent and stuff but like that's the one challenge but yeah i i i i completely agree but if they could fix that like that would be epic because then you could just ship so much faster but now going back and forth make hey i want you to change this headline i want you to change that like that does take multiple rounds of revisions normally you know if you get a designer you know like an amazing one they would get closer on the first shot yeah but if you're a junior designer there's a lot of rounds of feedback yeah it's like you're going back and forth anyway And a lot of this.

42:44So like, for example, you know, this headline was, I'll call it benefit driven. It pulled in one that was more feature right here. And so you just say, hey, you just give it feedback. Hey, Reneth, I want it to be a benefit headline. You pulled in a feature and there's not enough spacing. Please make those changes. Please save this so you don't make this mistake again. Right. So each time you do it, it's just learning your preferences more and more. So you posted this video. And so it uploaded that ad. oh you even tried to get it to whitelist me yeah yeah and then it told us hey you don't have permissions you know here's here's what to do but so it it launched that um you know creative analysis so this is alicia from motion and so analyzing the ad yeah you know can do a really great job of that um oh this is a pretty cool one caption so we launched this ad from tribe was doing pretty well it was b-roll not a talking head so i thought it needed captions uh um so run of added the captions to the video so it downloaded the original video from tribe using the api this is awesome um told it who it was you know created the naming conventions for it what campaign it lives in added the captions picked what landing page to go to um showed them here you know i didn't love there was something i didn't love i forget what it was but i it was like the format or a style of it.

44:06So I asked it to change that. Sure. Oh, I wanted to add then like a hook up top. Yeah. And so I didn't love how it did it. It came in late and whatever. And so I asked it to change that. And then I said, hey, now anytime I ask you to add a hook, let's do it this way, essentially. If you're a D2C brand not taking TikTok shop seriously yet, then you need to be. That's where T-Bar Partners comes in. They're the actual operators behind some of the biggest TikTok shop brands right now. Ridge, HexCloud, Groons, most importantly, us at Winx and more. T-Bar built the system behind all of them. Creator sourcing, sampling strategy, Discord management.

44:47They help building a real loyal TikTok shop community. And a huge bonus is they give brands a ton of high-performing UGC content that crushes it on meta. So if you're looking to scale up TikTok shop, visit tbarpartners.com.

45:00Matt Bertulli: So it's edit. This is like for people listening, literally editing video in slack right now it's adding captions to it yeah but still that that has to edit video yeah and the next step is like to cut it up yeah combine this see i use cloud code for that oh really like i open up a cloud code session uh and i edit the videos right in there okay and we have um skills especially any animation inside of uh yeah yeah for that and it's like it's pretty good i took what sean built and i'm just like okay that's good like Like, how do we make it as repeatable as possible? So you can now go in there and say, like, we have consistent characters.

45:34You can animate anything for the brand. So, yeah, video editing is, like, very close. So, like, but, I mean, again, think about, you know, we had four or five editors, and I wasn't doing any design myself, any editing. Like, five years ago, I tried to teach myself editing, you know, but now, like, are we going to be the, you and I going to be the best editor in the world? No, but, like, a basic UGC thing, we can do things like that. And so I think that's what's so cool is, like, you know, if you have a marketing team and you have a growth marketer, like they can have this idea and then be empowered to go and do it and not need to put it in Asana and do a brief and get three different people reviewing it and then go and do it.

46:11Matt Bertulli: What's your take on this whole like ad volume versus quality of ads? Like, and I get that, you know, to make great ads, you got to make a lot of ads. but it seems like everything we're talking about it would be easy for somebody to look at this and just be like great you guys are promoting making like a 5 000 sloppy ads every day where is this going like i actually wonder is this going to force us all to make better ads because everybody can make these now i think quality first and then volume i feel like everyone you ask has a different you know opinion of this um i mean if you're doing the same ad and same messaging you know you're not going to get a better hit rate or a better win rate so i think get the performance get quality learn what works and then you can just ship slop and it can work like sure i was talking to some some e-com drop shipping guy in dms the other day dude it's asking me some questions and i was asking him and he's like yeah it's just ai slop bro but like i think he's getting some i think he's getting some winners first though but no when when we reached out to sean the other day to like get some ideas on offers and stuff the first thing he said was let's just launch a thousand ads yeah i know it's like sean we want internet offer he's like launch a thousand ads like that's not what we asked for but like ima and stuff like that like you can say they're they're all launching a ton of ads you know it's mental but i think they're getting winners and it's just like how do you compete with that like if you're if you're not and you're doing things the old way like i don't know how you yeah there are ad accounts i've there's one i don't want to name names uh in one of my chat groups there's 14 000 active ads yeah in that ad account yeah and i'm like i get it they're big but like that's a lot of ads now i think 90 of it is ai slop but i don't think they would have them in there if they weren't working yeah like at that scale so there's also part of me thinking like the old school guy i mean is like just make great pieces of content and then the other part of me is like well it's working you know people seem to be reacting to this crap so yeah i think there's some bar level of quality and i know that there are people that say that the way you get quality is actually volume yeah and i i tend to be in that camp you know just make a lot and you'll get better at it yeah yeah yeah as long as you're actually getting better at it i think like if you're working with creators and we're on tribe like it is a volume thing but you do have to then learn from it and get better at it you can't just make the same and we have to coach them and we have to like do all that stuff yeah okay so uh hq omni send rich panel all of our site work there's higgs field malleus runneth because all software now has an element of ai built into it um so like mike is doing a ton of modeling right so like uh he took a lot of what he built at simple modern around like taking in real-time sales data and And can we build a financial model for the business that is updating like in the minute as orders come in?

49:00Matt Bertulli: It's then building like individual customer and individual SKU forecasting, which I think is a very interesting use case for AI. All of our asset management. Do you want to show people Winx OS and how we're doing this? Yeah, definitely. So we did this at Jones Road and I just wanted to set it up. So I set this up before launching, just like kind of very similar HQ. Like anything that is in WinxOS also lives in HQ. Yep. But it's just like, how do we get everything in one place? And how do you get a human viewable version of what's in HQ? Fair. That's how I like I use WinxOS. It's like, I still need to be able to look at this stuff sometimes.

49:38I mean, here's a good example, like a design system. Right. Like what lives in GitHub is just markdown files.

49:45Matt Bertulli: Right. And CSS. Yeah. And it turns out we suck at reading those. Yeah. And so this is just, you know, a way to visualize it. and then we have a shareable asset you know a way to share this but it's just here are our rules for typography and font and when we do and use certain things yep um just so that they you know stay consistent so things like that are in here um also like again like anytime we save an ad right like i i have them in here like and i just have like different you know landers let me see if i could find, you know, some of them, but like just different ads that we've saved. Sure.

50:24Because, you know, if you're using all of these different tools, then your agents don't might not have access to all of them. Yeah. And so, you know, that's that's another example of it.

50:35Matt Bertulli: Subscription brands, I am talking to you. You obsess over churn, but you only know your subscribers by skew, by cadence. and the real cancellation risk, it lies in the actual humans behind that data. Outer signal enriches your existing customer data, the existing Shopify list, and it turns it into real identifiable people and personas. We are talking demographics, lifestyle, who's a VIP, who's a creator, who is buying for a business. All of these persona attributes then get pushed into your favorite email platform so that your retention and win back flows can speak to a segment, a specific small group of people instead of everybody on a 30-day plan.

51:12Matt Bertulli: There's a native Shopify install. It's set up in minutes. Get your free persona report on your customer data. When you sign up, go to outerSignal.com slash operators. Tell them we sent you. I promise it'll help. And then Asset Library, same thing. It was like, okay, how do we get it? So I never want to go to an agent and be like, hey, let me go and I'm going to upload this image and use this image. It just should know. And so in Creative, we built an asset library uh like one of the things you can do for people again listening watching is you can set up uh a process like if you're a doom scroller like i'm a doom scroller and i like to just find stuff uh all i do is like bookmark or save it and then my agent picks that up and we'll have your harmonies do that yeah okay cool yeah i actually just had um muse do it because muse you know is the instagram one yeah it's built in yeah but yeah so wait so does muse have access to your instagram Yeah.

52:10Matt Bertulli: Yeah. Yeah. Like I showed my wife cause we're building a house right now and my wife is doing a lot of it. And so I set her up on her muse the other day and I was like, Hey, go, go ask it to find these fixtures and you're, and you're saved. Yeah. So that's like a really cool. Oh yeah. Cause I think that's like one of the downsides to a lot of this stuff is, um, Instagram's a closed playground, right? Like you got to use different tools to scrape and all that. And it's really inefficient, like inefficient and they use a lot of tokens. It costs money. But if, if. Might be good for creator stuff.

52:38Matt Bertulli: Yeah. That's kind of what I'm thinking. Yeah. So yeah, so here's all our assets. Again, it's not perfect. I think I have to just do a little bit like architecture on it, but it has all of our assets. You can download them. They're all saved here. And I actually built an MCP for this asset library. So now it can pull from them and not have to use the browser, but just wanted to get everything in one place versus having all of these different places and directions and things like that. Very cool. Before we finish, can you just show people? I'm going to show something. It'll take a little bit of time.

53:11I'm also going to show a more complicated version. This is kind of like, not to like to my own horn, but like something that is not like a 101. Okay. Just to show like what it can be potentially capable of because these things can be really, really powerful if you allow it to do that. Sweet. It's kind of pull a few of these principles together that we talked about. Okay. So number one, I'm going to use a skill. So I have this Winx page builder skill. and have trained the skill it's when i built for jones road and ported over to winx but it is going to build us a new landing page and it's kind of trained on analyzing an existing landing page figuring out what are the principles that makes that page successful and how do we build that page for winx using our design system okay so javi has great pages i don't think they'll mind that we're using ours they're like they're like one of the best landing pages they have there they're amazing pages i love their ux so we're going to pick theirs um and i'm a big whisper whisper flow guy.

54:06So I'll kind of, you know, yap a lot and do things like that. But this is going to use the skill. It's going to use context that we have about our business and our brand. And I'm going to set up what's called a loop or goal. I'm actually going to spin out some sub agents. This is where it's getting a little bit more complicated. Within here, I do have runneth with the CLI. I have access to it. This is Codex. I do have Claude and then Grok models as well. One thing agents are bad at is grading their own homework. Yeah. Right. And so sometimes it helps to bring them out. And I'm going to do a goal, which is essentially saying don't stop until you do this.

54:39Yep. That being said, we're doing design thing, which is subjective. And it is harder to do. It's called non-verifiable. And so I'm going to try to make it more objective and verifiable, but it's not always perfect. Okay. How are you going to make it more objective? I will. I'll show you. I would like you to make a new landing page for Winx using the page builder skill and this Javi page. I want you to use all of the context that you have in HQ about our business. Use our current or recent ad performance to take any learnings that you have from this. I want you to set up a goal. I want you to propose an overly ambitious goal.

55:16I want you to spin up adversarial sub-agents using Runeth, using Claude Code on Fable 5.1, Grok as well. I want you to create objective criteria where you're going to spin up creative director sub agents you're going to do at least three rounds of revisions and you are going to get it into a 9.5 out of 10 until you think i would be happy shipping this page live ask me any questions you have set an ambitious goal and then start boom and now it'll work and we don't have to sit here we'll say it well it's

55:55Matt Bertulli: going to go do work yeah and it'll probably come back and ask you questions because that's what it does but then once it's going it's gonna it's just gonna work for a while yeah right so we're not gonna make people sit here and watch because like that sort of thing and and do you always do this on like astra this is a nerdy question astra medium like why why medium astra is a new model and so it's just very powerful uh very intelligent you know it's their most powerful model so it will use more tokens yeah and so i think medium is just it's it's a good mix where it can do most things don't put on extra yeah i or or else it would burn your weekly you know if you say hello to astro on extra high it's like there goes half your week okay so while it's working i think let's let's wrap and i want to say this everything ai is changing all the damn time like every week is a massive like massive whiplash sort of model new models then like this one-upmanship and it's like i mean these labs just continually punch each other in the face and it's fun to watch as a spectator but i think if you try to keep up with it on a daily weekly basis it's it's a lot yeah so if i could leave people with sort of i think there's three things that i think you absolutely need to like get in place right now um so the first is and we started off the episode with like company context so this is like shared memory inherent knowledge as you build a business you're collecting this stuff you got to put it all in one place it needs to be agent friendly and it needs to be human friendly right would you agree that's number one definitely okay number two and this is uh i have a very strong bias here because i've seen the value of it at this point is you should build your own this is not company your own ceo founder brain and agent to take care of your brain everything that you read write watch anything you create every email you get everything that is like literally entering just your brain, not your teammates, just yours, that should all be stored and duplicated somewhere that is like AI friendly.

57:54Matt Bertulli: So that as you go through life, you're keeping a perfect copy of everything that you do with work and personal. So that like when I start my next day, I have like a perfect, like, here's what you're doing today. Here's context from the last week. I can see that you're meeting with Cody. This is the last thing that you talked to Cody about, like all of it's in my brain. And I say this because as a founder or a builder, a creator, whatever you want to call yourself, there is no way that you remember all of this. Like there's just not. I don't remember what I did last year on this day. I don't. My brain does though.

58:27Matt Bertulli: Like it has a perfect recounting of that. So I think that's number two. Would you agree? I think so. Yeah. Okay. I mean, I have it. I like, I use it as my note taker. Like, I don't use it. I'll just like whisper flow to my Hermes and be like, hey, I parked here. Yeah, yeah, yeah. Just keep everything in there. Okay, number three is what you just showed us, which is learn to work in goals and loops. So a lot of people use these tools and it's a very like linear and transactional. Hey, ChatGPT, hey, Claude, I would like you to go do this for me. And then they wait for it. And if they're crazy, then maybe they have like four or five or 10 of them going all at once.

59:06Matt Bertulli: But what you just showed us is like, I'm going to give it a job. I have a goal in mind. I'm going to give it a specific set of instructions. And you're just human language, this thing. And then you're going to let it go off and work. And it's going to figure out how to reach the objective. And it's going to loop around and do whatever. It's absolutely token heavy. Like it can be very token heavy. Would you agree with that? That that is like a skill that people working with AI, they should have that right now. Yeah. I would say first start using agents. And so I think only 2 % of the population are using agents.

59:41And most people are using chat. There are a lot of people who don't know what it can do. And there's people that are using it like a chat bot. And then there's people who are running entire businesses and their whole lives with it. And I think agents are becoming more popular as, you know, Muse or Grokbot or Instinct are, you know, making it easier to set up. But I would say just start experimenting. If you're using it like that, like start somewhere and start having it do things for you and taking things that you used to do manually. And it could be as easy as organizing your inbox. It's a great place to start.

1:00:08Or organizing your files on your desktop. But as far as go and do this action for me, buy this thing for me, do this flight, upload these ads, take some manual things. And then once you're done with manual things, then take things that are actually, hopefully income producing, revenue producing, and create these ads, build these things. So just test that. And then once you get that, then putting it all together where you have your context, your skills, your MCPs, your software, your brain, agents, set up loops for that. And that can, you get the full power of these systems. Yeah, I like that. That's great framing.

1:00:44Awesome, dude. I think that's the pod.

From the publisher

“They’re the most capable ever, but not trained on your business.”

What pillars should every AI-first brand build, and which one comes first? 

Matt Bertulli (Founder, Pela Case & Lomi) and Cody Plofker (Advisor & fmr. CEO, Jones Road Beauty) open up the playbook behind Winks, a brand where AI does the grunt work and humans make the calls. Piece by piece, they break down the four components that make that split work. 

1. Company brain: Context, memory, skills 

2. Operating system (OS): Human interface 

3. Software and tools: Existing tech, SaaS 

4. Agents: Researcher, writer, analyst, etc. 

The era of “prompt engineering” is over. Company context lets agents upload ads and match landing pages without being told twice. Yet AI tools still misread research, which leaves one question: How much can you trust them? 

Powered By 

Richpanel 

https://9ops.co/richpanel 

Runneth 

https://9ops.co/motion-runneth 

Skio 

https://9ops.co/skio-9operators 

Omnisend 

https://9ops.co/omnisend 

TBAR Partners 

https://9ops.co/tbarpartners 

OuterSignal 

https://9ops.co/outersignal 

Operators Portal 

https://portal.9operators.com/dashboard 

Operators Newsletter 

https://9operators.com/ 

More from Operators

All episodes